New technology uses near-infrared imaging and machine learning to find hidden tumors

Tumors, even benign, can damage surrounding blood vessels and tissues. When they are malignant, they are aggressive and sneaky, often causing irreparable damage. In the latter case, early detection is the key to treatment and recovery. However, such detection may require more advanced imaging techniques than those on the market today.

For example, some tumors develop deep into organs and tissues covered by a lamina propria, so standard methods such as endoscopy (a small camera is inserted into the patient’s body through a thin tube). It is difficult to observe and reach directly with. Those during biopsy. In particular, gastrointestinal stromal tumors (GIST) (usually found in the stomach and small intestine) are very time consuming and require rigorous techniques to prolong the diagnosis.

Now, to improve the GIST diagnosis, Dr. Taiki Sato, Hiroaki Ikematsu, Takeshi Kuwata, Dr. Hideo Yokota, Ph.D., National Cancer Center Hospital. Toshihiro Takamatsu and Kohei Soga of Tokyo University of Science, led by Dr. Hiroshi Takemura, have developed a technology that uses near-infrared hyperspectral imaging (NIR-HSI) along with machine learning.Their findings are published in Nature’s Science report..

This technique is a bit like X-rays. In other words, the idea is to use electromagnetic radiation that can pass through the body to create an image of the internal structure. The difference is that X-rays are at 0.01-10 nm, while near-infrared rays are at about 800-2500 nm. At that wavelength, near-infrared rays make the tissue appear transparent on the image. And these wavelengths are less harmful to the patient than visible light. “

Takeshi Kuwata National Cancer Center East

This should mean that scientists can safely investigate what is hidden within the tissue, but even studies by Dr. Takemura and his colleagues have tried to use NIR-HSI for deep tumors like GIST. No one did. In honor of the late professor who started the journey, Dr. Takemura said, “This project was made possible thanks to the late professor Kazuhiro Kaneko, who broke the barrier between doctors and engineers, and established this collaboration. . We obey his wishes. “

Dr. Takemura’s team conducted imaging experiments on 12 patients with confirmed cases of GIST whose tumors were surgically resected. Scientists used NIR-HSI to image the resected tissue and then had a pathologist examine the image to determine the boundary between normal and tumor tissue. We then used these images as training data for machine learning algorithms and basically instructed a computer program to distinguish between pixels in images representing normal tissue and images representing tumor tissue.

Scientists have found that machine learning analysis is 86% accurate in GIST identification, tumor and non-tumor sections, even though 10 of the 12 test tumors are completely or partially covered by the lamina propria. I found it to be effective for accurate color coding. “This is a very exciting development,” explains Dr. Takemura. “It is much easier for both patients and patients to be able to accurately, quickly and non-invasively diagnose different types of submucosal tumors without a biopsy, a procedure that requires surgery. Doctors.”

Dr. Takemura admits that there are still challenges, but feels ready to resolve them. Researchers have identified several areas of improvement in results, such as significantly increasing the training dataset, adding information about tumor depth in machine learning algorithms, and including other types of tumors in the analysis. .. Work is also underway to develop a NIR-HSI system built on top of existing endoscopic technology.

“We have already built a device to attach the NIR-HSI camera to the edge of the endoscope and would like to perform NIR-HSI analysis directly on the patient, not just the surgically resected tissue. “In the future, this will help isolate GIST from other types of submucosal tumors that may be more malignant and dangerous. This study is made possible by this interdisciplinary collaboration.” This is the first step towards a far more groundbreaking study in the future. “